During October 2025, T.B.L. refactored the Dram KV Inference API within the pytorch/FBGEMM repository, focusing on improving the API’s flexibility and integration reliability. By abstracting the existing dram_kv APIs into a dedicated interface with more generic method names, T.B.L. enabled easier future integration and cross-team collaboration. The work also introduced synchronous operations to the inference embedding, enhancing determinism and stability for downstream users. This effort leveraged C++ development skills and expertise in API design and embedded systems. The changes were thoroughly documented and linked to relevant commits and pull requests, reflecting a methodical and maintainable engineering approach.